Deciphering the Effect of Sulfide Derivatives on the Prediction of Nitrite Accumulation in Sulfide-Dosed Partial Nitrification Using Machine Learning

Abstract

The unstable nature of partial nitrification (PN) has made it difficult to achieve stable nitrite production without robust process control. Sulfide has been reported to be a potential mediator for the achievement of PN by creating an inhibiting environment for nitrite-oxidizing bacteria (NOB). In the field sulfide salt, sodium sulfide (Na2S) increases pH due to the production of hydroxide ions that can inhibit NOBs. Because of the complexity in metabolic pathways and the presence of sulfide in ionized and unionized forms, conventional first principles models have limitations in providing accurate predictions. This study demonstrated a comparative analysis of three machine learning (ML) models to determine the most influential parameter for nitrite accumulation during sulfide addition. pH, HS−/N, H2S/N were selected as input parameters. The data from the lab-scale experiments were used for training of ML algorithms, namely Gaussian process regression (GPR), support vector machines (SVM), and ensemble regression tree (ER). The results showed GPR to be a better performer in prediction highlighting its advantage over other ML models with R2=0.95, RMSE=0.19 and MAE=0.14, and ionized form of sulfide (HS−/N) was found to be the significant parameter for the successful nitrite accumulation. This study highlights the integration of ML techniques to predict nitrite accumulation ratio (NAR) in real-world wastewater treatment applications by demonstrating a practical and impactful approach to optimize biological nitrogen removal processes, addressing both operational challenges and environmental concerns effectively.